lnet/selftest/framework.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lnet/selftest/framework.c
  37  *
  38  * Author: Isaac Huang <isaac@clusterfs.com>
  39  * Author: Liang Zhen  <liangzhen@clusterfs.com>
  40  */
  41
  42 #define DEBUG_SUBSYSTEM S_LNET
  43
  44 #include "selftest.h"
  45
  46 lst_sid_t LST_INVALID_SID = {LNET_NID_ANY, -1};
  47
  48 static int session_timeout = 100;
  49 CFS_MODULE_PARM(session_timeout, "i", int, 0444,
  50                 "test session timeout in seconds (100 by default, 0 == never)");
  51
  52 static int rpc_timeout = 64;
  53 CFS_MODULE_PARM(rpc_timeout, "i", int, 0644,
  54                 "rpc timeout in seconds (64 by default, 0 == never)");
  55
  56 #define sfw_unpack_id(id)               \
  57 do {                                    \
  58         __swab64s(&(id).nid);           \
  59         __swab32s(&(id).pid);           \
  60 } while (0)
  61
  62 #define sfw_unpack_sid(sid)             \
  63 do {                                    \
  64         __swab64s(&(sid).ses_nid);      \
  65         __swab64s(&(sid).ses_stamp);    \
  66 } while (0)
  67
  68 #define sfw_unpack_fw_counters(fc)        \
  69 do {                                      \
  70         __swab32s(&(fc).running_ms);      \
  71         __swab32s(&(fc).active_batches);  \
  72         __swab32s(&(fc).zombie_sessions); \
  73         __swab32s(&(fc).brw_errors);      \
  74         __swab32s(&(fc).ping_errors);     \
  75 } while (0)
  76
  77 #define sfw_unpack_rpc_counters(rc)     \
  78 do {                                    \
  79         __swab32s(&(rc).errors);        \
  80         __swab32s(&(rc).rpcs_sent);     \
  81         __swab32s(&(rc).rpcs_rcvd);     \
  82         __swab32s(&(rc).rpcs_dropped);  \
  83         __swab32s(&(rc).rpcs_expired);  \
  84         __swab64s(&(rc).bulk_get);      \
  85         __swab64s(&(rc).bulk_put);      \
  86 } while (0)
  87
  88 #define sfw_unpack_lnet_counters(lc)    \
  89 do {                                    \
  90         __swab32s(&(lc).errors);        \
  91         __swab32s(&(lc).msgs_max);      \
  92         __swab32s(&(lc).msgs_alloc);    \
  93         __swab32s(&(lc).send_count);    \
  94         __swab32s(&(lc).recv_count);    \
  95         __swab32s(&(lc).drop_count);    \
  96         __swab32s(&(lc).route_count);   \
  97         __swab64s(&(lc).send_length);   \
  98         __swab64s(&(lc).recv_length);   \
  99         __swab64s(&(lc).drop_length);   \
 100         __swab64s(&(lc).route_length);  \
 101 } while (0)
 102
 103 #define sfw_test_active(t)      (atomic_read(&(t)->tsi_nactive) != 0)
 104 #define sfw_batch_active(b)     (atomic_read(&(b)->bat_nactive) != 0)
 105
 106 struct smoketest_framework {
 107         /* RPCs to be recycled */
 108         struct list_head        fw_zombie_rpcs;
 109         /* stopping sessions */
 110         struct list_head        fw_zombie_sessions;
 111         /* registered test cases */
 112         struct list_head        fw_tests;
 113         /* # zombie sessions */
 114         atomic_t                fw_nzombies;
 115         /* serialise */
 116         spinlock_t              fw_lock;
 117         /* _the_ session */
 118         sfw_session_t           *fw_session;
 119         /* shutdown in progress */
 120         int                     fw_shuttingdown;
 121         /* running RPC */
 122         srpc_server_rpc_t       *fw_active_srpc;
 123 } sfw_data;
 124
 125 /* forward ref's */
 126 int sfw_stop_batch (sfw_batch_t *tsb, int force);
 127 void sfw_destroy_session (sfw_session_t *sn);
 128
 129 static inline sfw_test_case_t *
 130 sfw_find_test_case(int id)
 131 {
 132         sfw_test_case_t *tsc;
 133
 134         LASSERT(id <= SRPC_SERVICE_MAX_ID);
 135         LASSERT(id > SRPC_FRAMEWORK_SERVICE_MAX_ID);
 136
 137         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
 138                 if (tsc->tsc_srv_service->sv_id == id)
 139                         return tsc;
 140         }
 141
 142         return NULL;
 143 }
 144
 145 static int
 146 sfw_register_test (srpc_service_t *service, sfw_test_client_ops_t *cliops)
 147 {
 148         sfw_test_case_t *tsc;
 149
 150         if (sfw_find_test_case(service->sv_id) != NULL) {
 151                 CERROR ("Failed to register test %s (%d)\n",
 152                         service->sv_name, service->sv_id);
 153                 return -EEXIST;
 154         }
 155
 156         LIBCFS_ALLOC(tsc, sizeof(sfw_test_case_t));
 157         if (tsc == NULL)
 158                 return -ENOMEM;
 159
 160         memset(tsc, 0, sizeof(sfw_test_case_t));
 161         tsc->tsc_cli_ops     = cliops;
 162         tsc->tsc_srv_service = service;
 163
 164         list_add_tail(&tsc->tsc_list, &sfw_data.fw_tests);
 165         return 0;
 166 }
 167
 168 void
 169 sfw_add_session_timer (void)
 170 {
 171         sfw_session_t *sn = sfw_data.fw_session;
 172         stt_timer_t   *timer = &sn->sn_timer;
 173
 174         LASSERT (!sfw_data.fw_shuttingdown);
 175
 176         if (sn == NULL || sn->sn_timeout == 0)
 177                 return;
 178
 179         LASSERT (!sn->sn_timer_active);
 180
 181         sn->sn_timer_active = 1;
 182         timer->stt_expires = cfs_time_add(sn->sn_timeout,
 183                                           cfs_time_current_sec());
 184         stt_add_timer(timer);
 185         return;
 186 }
 187
 188 int
 189 sfw_del_session_timer (void)
 190 {
 191         sfw_session_t *sn = sfw_data.fw_session;
 192
 193         if (sn == NULL || !sn->sn_timer_active)
 194                 return 0;
 195
 196         LASSERT (sn->sn_timeout != 0);
 197
 198         if (stt_del_timer(&sn->sn_timer)) { /* timer defused */
 199                 sn->sn_timer_active = 0;
 200                 return 0;
 201         }
 202
 203 #ifndef __KERNEL__
 204         /* Racing is impossible in single-threaded userland selftest */
 205         LBUG();
 206 #endif
 207         return EBUSY; /* racing with sfw_session_expired() */
 208 }
 209
 210 /* called with sfw_data.fw_lock held */
 211 static void
 212 sfw_deactivate_session (void)
 213 {
 214         sfw_session_t *sn = sfw_data.fw_session;
 215         int            nactive = 0;
 216         sfw_batch_t   *tsb;
 217         sfw_test_case_t *tsc;
 218
 219         if (sn == NULL) return;
 220
 221         LASSERT(!sn->sn_timer_active);
 222
 223         sfw_data.fw_session = NULL;
 224         atomic_inc(&sfw_data.fw_nzombies);
 225         list_add(&sn->sn_list, &sfw_data.fw_zombie_sessions);
 226
 227         spin_unlock(&sfw_data.fw_lock);
 228
 229         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
 230                 srpc_abort_service(tsc->tsc_srv_service);
 231         }
 232
 233         spin_lock(&sfw_data.fw_lock);
 234
 235         list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
 236                 if (sfw_batch_active(tsb)) {
 237                         nactive++;
 238                         sfw_stop_batch(tsb, 1);
 239                 }
 240         }
 241
 242         if (nactive != 0)
 243                 return; /* wait for active batches to stop */
 244
 245         list_del_init(&sn->sn_list);
 246         spin_unlock(&sfw_data.fw_lock);
 247
 248         sfw_destroy_session(sn);
 249
 250         spin_lock(&sfw_data.fw_lock);
 251 }
 252
 253 #ifndef __KERNEL__
 254
 255 int
 256 sfw_session_removed (void)
 257 {
 258         return (sfw_data.fw_session == NULL) ? 1 : 0;
 259 }
 260
 261 #endif
 262
 263 void
 264 sfw_session_expired (void *data)
 265 {
 266         sfw_session_t *sn = data;
 267
 268         spin_lock(&sfw_data.fw_lock);
 269
 270         LASSERT (sn->sn_timer_active);
 271         LASSERT (sn == sfw_data.fw_session);
 272
 273         CWARN ("Session expired! sid: %s-"LPU64", name: %s\n",
 274                libcfs_nid2str(sn->sn_id.ses_nid),
 275                sn->sn_id.ses_stamp, &sn->sn_name[0]);
 276
 277         sn->sn_timer_active = 0;
 278         sfw_deactivate_session();
 279
 280         spin_unlock(&sfw_data.fw_lock);
 281 }
 282
 283 static inline void
 284 sfw_init_session(sfw_session_t *sn, lst_sid_t sid,
 285                  unsigned features, const char *name)
 286 {
 287         stt_timer_t *timer = &sn->sn_timer;
 288
 289         memset(sn, 0, sizeof(sfw_session_t));
 290         INIT_LIST_HEAD(&sn->sn_list);
 291         INIT_LIST_HEAD(&sn->sn_batches);
 292         atomic_set(&sn->sn_refcount, 1);        /* +1 for caller */
 293         atomic_set(&sn->sn_brw_errors, 0);
 294         atomic_set(&sn->sn_ping_errors, 0);
 295         strlcpy(&sn->sn_name[0], name, sizeof(sn->sn_name));
 296
 297         sn->sn_timer_active = 0;
 298         sn->sn_id           = sid;
 299         sn->sn_features     = features;
 300         sn->sn_timeout      = session_timeout;
 301         sn->sn_started      = cfs_time_current();
 302
 303         timer->stt_data = sn;
 304         timer->stt_func = sfw_session_expired;
 305         INIT_LIST_HEAD(&timer->stt_list);
 306 }
 307
 308 /* completion handler for incoming framework RPCs */
 309 void
 310 sfw_server_rpc_done(struct srpc_server_rpc *rpc)
 311 {
 312         struct srpc_service     *sv     = rpc->srpc_scd->scd_svc;
 313         int                     status  = rpc->srpc_status;
 314
 315         CDEBUG (D_NET,
 316                 "Incoming framework RPC done: "
 317                 "service %s, peer %s, status %s:%d\n",
 318                 sv->sv_name, libcfs_id2str(rpc->srpc_peer),
 319                 swi_state2str(rpc->srpc_wi.swi_state),
 320                 status);
 321
 322         if (rpc->srpc_bulk != NULL)
 323                 sfw_free_pages(rpc);
 324         return;
 325 }
 326
 327 void
 328 sfw_client_rpc_fini (srpc_client_rpc_t *rpc)
 329 {
 330         LASSERT(rpc->crpc_bulk.bk_niov == 0);
 331         LASSERT(list_empty(&rpc->crpc_list));
 332         LASSERT(atomic_read(&rpc->crpc_refcount) == 0);
 333 #ifndef __KERNEL__
 334         LASSERT(rpc->crpc_bulk.bk_pages == NULL);
 335 #endif
 336
 337         CDEBUG(D_NET, "Outgoing framework RPC done: "
 338                "service %d, peer %s, status %s:%d:%d\n",
 339                rpc->crpc_service, libcfs_id2str(rpc->crpc_dest),
 340                swi_state2str(rpc->crpc_wi.swi_state),
 341                rpc->crpc_aborted, rpc->crpc_status);
 342
 343         spin_lock(&sfw_data.fw_lock);
 344
 345         /* my callers must finish all RPCs before shutting me down */
 346         LASSERT(!sfw_data.fw_shuttingdown);
 347         list_add(&rpc->crpc_list, &sfw_data.fw_zombie_rpcs);
 348
 349         spin_unlock(&sfw_data.fw_lock);
 350 }
 351
 352 sfw_batch_t *
 353 sfw_find_batch (lst_bid_t bid)
 354 {
 355         sfw_session_t *sn = sfw_data.fw_session;
 356         sfw_batch_t   *bat;
 357
 358         LASSERT(sn != NULL);
 359
 360         list_for_each_entry(bat, &sn->sn_batches, bat_list) {
 361                 if (bat->bat_id.bat_id == bid.bat_id)
 362                         return bat;
 363         }
 364
 365         return NULL;
 366 }
 367
 368 sfw_batch_t *
 369 sfw_bid2batch (lst_bid_t bid)
 370 {
 371         sfw_session_t *sn = sfw_data.fw_session;
 372         sfw_batch_t   *bat;
 373
 374         LASSERT (sn != NULL);
 375
 376         bat = sfw_find_batch(bid);
 377         if (bat != NULL)
 378                 return bat;
 379
 380         LIBCFS_ALLOC(bat, sizeof(sfw_batch_t));
 381         if (bat == NULL)
 382                 return NULL;
 383
 384         bat->bat_error   = 0;
 385         bat->bat_session = sn;
 386         bat->bat_id      = bid;
 387         atomic_set(&bat->bat_nactive, 0);
 388         INIT_LIST_HEAD(&bat->bat_tests);
 389
 390         list_add_tail(&bat->bat_list, &sn->sn_batches);
 391         return bat;
 392 }
 393
 394 int
 395 sfw_get_stats (srpc_stat_reqst_t *request, srpc_stat_reply_t *reply)
 396 {
 397         sfw_session_t  *sn = sfw_data.fw_session;
 398         sfw_counters_t *cnt = &reply->str_fw;
 399         sfw_batch_t    *bat;
 400         struct timeval  tv;
 401
 402         reply->str_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 403
 404         if (request->str_sid.ses_nid == LNET_NID_ANY) {
 405                 reply->str_status = EINVAL;
 406                 return 0;
 407         }
 408
 409         if (sn == NULL || !sfw_sid_equal(request->str_sid, sn->sn_id)) {
 410                 reply->str_status = ESRCH;
 411                 return 0;
 412         }
 413
 414         lnet_counters_get(&reply->str_lnet);
 415         srpc_get_counters(&reply->str_rpc);
 416
 417         /* send over the msecs since the session was started
 418          - with 32 bits to send, this is ~49 days */
 419         cfs_duration_usec(cfs_time_sub(cfs_time_current(),
 420                                        sn->sn_started), &tv);
 421
 422         cnt->running_ms      = (__u32)(tv.tv_sec * 1000 + tv.tv_usec / 1000);
 423         cnt->brw_errors      = atomic_read(&sn->sn_brw_errors);
 424         cnt->ping_errors     = atomic_read(&sn->sn_ping_errors);
 425         cnt->zombie_sessions = atomic_read(&sfw_data.fw_nzombies);
 426
 427         cnt->active_batches = 0;
 428         list_for_each_entry(bat, &sn->sn_batches, bat_list) {
 429                 if (atomic_read(&bat->bat_nactive) > 0)
 430                         cnt->active_batches++;
 431         }
 432
 433         reply->str_status = 0;
 434         return 0;
 435 }
 436
 437 int
 438 sfw_make_session(srpc_mksn_reqst_t *request, srpc_mksn_reply_t *reply)
 439 {
 440         sfw_session_t *sn = sfw_data.fw_session;
 441         srpc_msg_t    *msg = container_of(request, srpc_msg_t,
 442                                           msg_body.mksn_reqst);
 443         int            cplen = 0;
 444
 445         if (request->mksn_sid.ses_nid == LNET_NID_ANY) {
 446                 reply->mksn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 447                 reply->mksn_status = EINVAL;
 448                 return 0;
 449         }
 450
 451         if (sn != NULL) {
 452                 reply->mksn_status  = 0;
 453                 reply->mksn_sid     = sn->sn_id;
 454                 reply->mksn_timeout = sn->sn_timeout;
 455
 456                 if (sfw_sid_equal(request->mksn_sid, sn->sn_id)) {
 457                         atomic_inc(&sn->sn_refcount);
 458                         return 0;
 459                 }
 460
 461                 if (!request->mksn_force) {
 462                         reply->mksn_status = EBUSY;
 463                         cplen = strlcpy(&reply->mksn_name[0], &sn->sn_name[0],
 464                                         sizeof(reply->mksn_name));
 465                         if (cplen >= sizeof(reply->mksn_name))
 466                                 return -E2BIG;
 467                         return 0;
 468                 }
 469         }
 470
 471         /* reject the request if it requires unknown features
 472          * NB: old version will always accept all features because it's not
 473          * aware of srpc_msg_t::msg_ses_feats, it's a defect but it's also
 474          * harmless because it will return zero feature to console, and it's
 475          * console's responsibility to make sure all nodes in a session have
 476          * same feature mask. */
 477         if ((msg->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
 478                 reply->mksn_status = EPROTO;
 479                 return 0;
 480         }
 481
 482         /* brand new or create by force */
 483         LIBCFS_ALLOC(sn, sizeof(sfw_session_t));
 484         if (sn == NULL) {
 485                 CERROR("dropping RPC mksn under memory pressure\n");
 486                 return -ENOMEM;
 487         }
 488
 489         sfw_init_session(sn, request->mksn_sid,
 490                          msg->msg_ses_feats, &request->mksn_name[0]);
 491
 492         spin_lock(&sfw_data.fw_lock);
 493
 494         sfw_deactivate_session();
 495         LASSERT(sfw_data.fw_session == NULL);
 496         sfw_data.fw_session = sn;
 497
 498         spin_unlock(&sfw_data.fw_lock);
 499
 500         reply->mksn_status  = 0;
 501         reply->mksn_sid     = sn->sn_id;
 502         reply->mksn_timeout = sn->sn_timeout;
 503         return 0;
 504 }
 505
 506 int
 507 sfw_remove_session (srpc_rmsn_reqst_t *request, srpc_rmsn_reply_t *reply)
 508 {
 509         sfw_session_t *sn = sfw_data.fw_session;
 510
 511         reply->rmsn_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
 512
 513         if (request->rmsn_sid.ses_nid == LNET_NID_ANY) {
 514                 reply->rmsn_status = EINVAL;
 515                 return 0;
 516         }
 517
 518         if (sn == NULL || !sfw_sid_equal(request->rmsn_sid, sn->sn_id)) {
 519                 reply->rmsn_status = (sn == NULL) ? ESRCH : EBUSY;
 520                 return 0;
 521         }
 522
 523         if (!atomic_dec_and_test(&sn->sn_refcount)) {
 524                 reply->rmsn_status = 0;
 525                 return 0;
 526         }
 527
 528         spin_lock(&sfw_data.fw_lock);
 529         sfw_deactivate_session();
 530         spin_unlock(&sfw_data.fw_lock);
 531
 532         reply->rmsn_status = 0;
 533         reply->rmsn_sid    = LST_INVALID_SID;
 534         LASSERT(sfw_data.fw_session == NULL);
 535         return 0;
 536 }
 537
 538 int
 539 sfw_debug_session (srpc_debug_reqst_t *request, srpc_debug_reply_t *reply)
 540 {
 541         sfw_session_t *sn = sfw_data.fw_session;
 542
 543         if (sn == NULL) {
 544                 reply->dbg_status = ESRCH;
 545                 reply->dbg_sid    = LST_INVALID_SID;
 546                 return 0;
 547         }
 548
 549         reply->dbg_status  = 0;
 550         reply->dbg_sid     = sn->sn_id;
 551         reply->dbg_timeout = sn->sn_timeout;
 552         if (strlcpy(reply->dbg_name, &sn->sn_name[0], sizeof(reply->dbg_name))
 553             >= sizeof(reply->dbg_name))
 554                 return -E2BIG;
 555
 556         return 0;
 557 }
 558
 559 void
 560 sfw_test_rpc_fini (srpc_client_rpc_t *rpc)
 561 {
 562         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 563         sfw_test_instance_t *tsi = tsu->tsu_instance;
 564
 565         /* Called with hold of tsi->tsi_lock */
 566         LASSERT(list_empty(&rpc->crpc_list));
 567         list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
 568 }
 569
 570 static inline int
 571 sfw_test_buffers(sfw_test_instance_t *tsi)
 572 {
 573         struct sfw_test_case    *tsc;
 574         struct srpc_service     *svc;
 575         int                     nbuf;
 576
 577         LASSERT(tsi != NULL);
 578         tsc = sfw_find_test_case(tsi->tsi_service);
 579         LASSERT(tsc != NULL);
 580         svc = tsc->tsc_srv_service;
 581         LASSERT(svc != NULL);
 582
 583         nbuf = min(svc->sv_wi_total, tsi->tsi_loop) / svc->sv_ncpts;
 584         return max(SFW_TEST_WI_MIN, nbuf + SFW_TEST_WI_EXTRA);
 585 }
 586
 587 int
 588 sfw_load_test(struct sfw_test_instance *tsi)
 589 {
 590         struct sfw_test_case    *tsc;
 591         struct srpc_service     *svc;
 592         int                     nbuf;
 593         int                     rc;
 594
 595         LASSERT(tsi != NULL);
 596         tsc = sfw_find_test_case(tsi->tsi_service);
 597         nbuf = sfw_test_buffers(tsi);
 598         LASSERT(tsc != NULL);
 599         svc = tsc->tsc_srv_service;
 600
 601         if (tsi->tsi_is_client) {
 602                 tsi->tsi_ops = tsc->tsc_cli_ops;
 603                 return 0;
 604         }
 605
 606         rc = srpc_service_add_buffers(svc, nbuf);
 607         if (rc != 0) {
 608                 CWARN("Failed to reserve enough buffers: "
 609                       "service %s, %d needed: %d\n", svc->sv_name, nbuf, rc);
 610                 /* NB: this error handler is not strictly correct, because
 611                  * it may release more buffers than already allocated,
 612                  * but it doesn't matter because request portal should
 613                  * be lazy portal and will grow buffers if necessary. */
 614                 srpc_service_remove_buffers(svc, nbuf);
 615                 return -ENOMEM;
 616         }
 617
 618         CDEBUG(D_NET, "Reserved %d buffers for test %s\n",
 619                nbuf * (srpc_serv_is_framework(svc) ?
 620                        1 : cfs_cpt_number(cfs_cpt_table)), svc->sv_name);
 621         return 0;
 622 }
 623
 624 void
 625 sfw_unload_test(struct sfw_test_instance *tsi)
 626 {
 627         struct sfw_test_case *tsc;
 628
 629         LASSERT(tsi != NULL);
 630         tsc = sfw_find_test_case(tsi->tsi_service);
 631         LASSERT(tsc != NULL);
 632
 633         if (tsi->tsi_is_client)
 634                 return;
 635
 636         /* shrink buffers, because request portal is lazy portal
 637          * which can grow buffers at runtime so we may leave
 638          * some buffers behind, but never mind... */
 639         srpc_service_remove_buffers(tsc->tsc_srv_service,
 640                                     sfw_test_buffers(tsi));
 641         return;
 642 }
 643
 644 void
 645 sfw_destroy_test_instance (sfw_test_instance_t *tsi)
 646 {
 647         srpc_client_rpc_t *rpc;
 648         sfw_test_unit_t   *tsu;
 649
 650         if (!tsi->tsi_is_client) goto clean;
 651
 652         tsi->tsi_ops->tso_fini(tsi);
 653
 654         LASSERT(!tsi->tsi_stopping);
 655         LASSERT(list_empty(&tsi->tsi_active_rpcs));
 656         LASSERT(!sfw_test_active(tsi));
 657
 658         while (!list_empty(&tsi->tsi_units)) {
 659                 tsu = list_entry(tsi->tsi_units.next,
 660                                  sfw_test_unit_t, tsu_list);
 661                 list_del(&tsu->tsu_list);
 662                 LIBCFS_FREE(tsu, sizeof(*tsu));
 663         }
 664
 665         while (!list_empty(&tsi->tsi_free_rpcs)) {
 666                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 667                                  srpc_client_rpc_t, crpc_list);
 668                 list_del(&rpc->crpc_list);
 669                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
 670         }
 671
 672 clean:
 673         sfw_unload_test(tsi);
 674         LIBCFS_FREE(tsi, sizeof(*tsi));
 675         return;
 676 }
 677
 678 void
 679 sfw_destroy_batch (sfw_batch_t *tsb)
 680 {
 681         sfw_test_instance_t *tsi;
 682
 683         LASSERT(!sfw_batch_active(tsb));
 684         LASSERT(list_empty(&tsb->bat_list));
 685
 686         while (!list_empty(&tsb->bat_tests)) {
 687                 tsi = list_entry(tsb->bat_tests.next,
 688                                  sfw_test_instance_t, tsi_list);
 689                 list_del_init(&tsi->tsi_list);
 690                 sfw_destroy_test_instance(tsi);
 691         }
 692
 693         LIBCFS_FREE(tsb, sizeof(sfw_batch_t));
 694         return;
 695 }
 696
 697 void
 698 sfw_destroy_session (sfw_session_t *sn)
 699 {
 700         sfw_batch_t *batch;
 701
 702         LASSERT(list_empty(&sn->sn_list));
 703         LASSERT(sn != sfw_data.fw_session);
 704
 705         while (!list_empty(&sn->sn_batches)) {
 706                 batch = list_entry(sn->sn_batches.next,
 707                                    sfw_batch_t, bat_list);
 708                 list_del_init(&batch->bat_list);
 709                 sfw_destroy_batch(batch);
 710         }
 711
 712         LIBCFS_FREE(sn, sizeof(*sn));
 713         atomic_dec(&sfw_data.fw_nzombies);
 714         return;
 715 }
 716
 717 void
 718 sfw_unpack_addtest_req(srpc_msg_t *msg)
 719 {
 720         srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
 721
 722         LASSERT (msg->msg_type == SRPC_MSG_TEST_REQST);
 723         LASSERT (req->tsr_is_client);
 724
 725         if (msg->msg_magic == SRPC_MSG_MAGIC)
 726                 return; /* no flipping needed */
 727
 728         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
 729
 730         if (req->tsr_service == SRPC_SERVICE_BRW) {
 731                 if ((msg->msg_ses_feats & LST_FEAT_BULK_LEN) == 0) {
 732                         test_bulk_req_t *bulk = &req->tsr_u.bulk_v0;
 733
 734                         __swab32s(&bulk->blk_opc);
 735                         __swab32s(&bulk->blk_npg);
 736                         __swab32s(&bulk->blk_flags);
 737
 738                 } else {
 739                         test_bulk_req_v1_t *bulk = &req->tsr_u.bulk_v1;
 740
 741                         __swab16s(&bulk->blk_opc);
 742                         __swab16s(&bulk->blk_flags);
 743                         __swab32s(&bulk->blk_offset);
 744                         __swab32s(&bulk->blk_len);
 745                 }
 746
 747                 return;
 748         }
 749
 750         if (req->tsr_service == SRPC_SERVICE_PING) {
 751                 test_ping_req_t *ping = &req->tsr_u.ping;
 752
 753                 __swab32s(&ping->png_size);
 754                 __swab32s(&ping->png_flags);
 755                 return;
 756         }
 757
 758         LBUG();
 759         return;
 760 }
 761
 762 int
 763 sfw_add_test_instance (sfw_batch_t *tsb, srpc_server_rpc_t *rpc)
 764 {
 765         srpc_msg_t          *msg = &rpc->srpc_reqstbuf->buf_msg;
 766         srpc_test_reqst_t   *req = &msg->msg_body.tes_reqst;
 767         srpc_bulk_t         *bk = rpc->srpc_bulk;
 768         int                  ndest = req->tsr_ndest;
 769         sfw_test_unit_t     *tsu;
 770         sfw_test_instance_t *tsi;
 771         int                  i;
 772         int                  rc;
 773
 774         LIBCFS_ALLOC(tsi, sizeof(*tsi));
 775         if (tsi == NULL) {
 776                 CERROR ("Can't allocate test instance for batch: "LPU64"\n",
 777                         tsb->bat_id.bat_id);
 778                 return -ENOMEM;
 779         }
 780
 781         memset(tsi, 0, sizeof(*tsi));
 782         spin_lock_init(&tsi->tsi_lock);
 783         atomic_set(&tsi->tsi_nactive, 0);
 784         INIT_LIST_HEAD(&tsi->tsi_units);
 785         INIT_LIST_HEAD(&tsi->tsi_free_rpcs);
 786         INIT_LIST_HEAD(&tsi->tsi_active_rpcs);
 787
 788         tsi->tsi_stopping      = 0;
 789         tsi->tsi_batch         = tsb;
 790         tsi->tsi_loop          = req->tsr_loop;
 791         tsi->tsi_concur        = req->tsr_concur;
 792         tsi->tsi_service       = req->tsr_service;
 793         tsi->tsi_is_client     = !!(req->tsr_is_client);
 794         tsi->tsi_stoptsu_onerr = !!(req->tsr_stop_onerr);
 795
 796         rc = sfw_load_test(tsi);
 797         if (rc != 0) {
 798                 LIBCFS_FREE(tsi, sizeof(*tsi));
 799                 return rc;
 800         }
 801
 802         LASSERT (!sfw_batch_active(tsb));
 803
 804         if (!tsi->tsi_is_client) {
 805                 /* it's test server, just add it to tsb */
 806                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 807                 return 0;
 808         }
 809
 810         LASSERT (bk != NULL);
 811 #ifndef __KERNEL__
 812         LASSERT (bk->bk_pages != NULL);
 813 #endif
 814         LASSERT (bk->bk_niov * SFW_ID_PER_PAGE >= (unsigned int)ndest);
 815         LASSERT((unsigned int)bk->bk_len >=
 816                 sizeof(lnet_process_id_packed_t) * ndest);
 817
 818         sfw_unpack_addtest_req(msg);
 819         memcpy(&tsi->tsi_u, &req->tsr_u, sizeof(tsi->tsi_u));
 820
 821         for (i = 0; i < ndest; i++) {
 822                 lnet_process_id_packed_t *dests;
 823                 lnet_process_id_packed_t  id;
 824                 int                       j;
 825
 826 #ifdef __KERNEL__
 827                 dests = page_address(bk->bk_iovs[i / SFW_ID_PER_PAGE].kiov_page);
 828                 LASSERT (dests != NULL);  /* my pages are within KVM always */
 829 #else
 830                 dests = page_address(bk->bk_pages[i / SFW_ID_PER_PAGE]);
 831 #endif
 832                 id = dests[i % SFW_ID_PER_PAGE];
 833                 if (msg->msg_magic != SRPC_MSG_MAGIC)
 834                         sfw_unpack_id(id);
 835
 836                 for (j = 0; j < tsi->tsi_concur; j++) {
 837                         LIBCFS_ALLOC(tsu, sizeof(sfw_test_unit_t));
 838                         if (tsu == NULL) {
 839                                 rc = -ENOMEM;
 840                                 CERROR ("Can't allocate tsu for %d\n",
 841                                         tsi->tsi_service);
 842                                 goto error;
 843                         }
 844
 845                         tsu->tsu_dest.nid = id.nid;
 846                         tsu->tsu_dest.pid = id.pid;
 847                         tsu->tsu_instance = tsi;
 848                         tsu->tsu_private  = NULL;
 849                         list_add_tail(&tsu->tsu_list, &tsi->tsi_units);
 850                 }
 851         }
 852
 853         rc = tsi->tsi_ops->tso_init(tsi);
 854         if (rc == 0) {
 855                 list_add_tail(&tsi->tsi_list, &tsb->bat_tests);
 856                 return 0;
 857         }
 858
 859 error:
 860         LASSERT(rc != 0);
 861         sfw_destroy_test_instance(tsi);
 862         return rc;
 863 }
 864
 865 static void
 866 sfw_test_unit_done (sfw_test_unit_t *tsu)
 867 {
 868         sfw_test_instance_t *tsi = tsu->tsu_instance;
 869         sfw_batch_t         *tsb = tsi->tsi_batch;
 870         sfw_session_t       *sn = tsb->bat_session;
 871
 872         LASSERT (sfw_test_active(tsi));
 873
 874         if (!atomic_dec_and_test(&tsi->tsi_nactive))
 875                 return;
 876
 877         /* the test instance is done */
 878         spin_lock(&tsi->tsi_lock);
 879
 880         tsi->tsi_stopping = 0;
 881
 882         spin_unlock(&tsi->tsi_lock);
 883
 884         spin_lock(&sfw_data.fw_lock);
 885
 886         if (!atomic_dec_and_test(&tsb->bat_nactive) ||/* tsb still active */
 887             sn == sfw_data.fw_session) {                  /* sn also active */
 888                 spin_unlock(&sfw_data.fw_lock);
 889                 return;
 890         }
 891
 892         LASSERT(!list_empty(&sn->sn_list)); /* I'm a zombie! */
 893
 894         list_for_each_entry(tsb, &sn->sn_batches, bat_list) {
 895                 if (sfw_batch_active(tsb)) {
 896                         spin_unlock(&sfw_data.fw_lock);
 897                         return;
 898                 }
 899         }
 900
 901         list_del_init(&sn->sn_list);
 902         spin_unlock(&sfw_data.fw_lock);
 903
 904         sfw_destroy_session(sn);
 905         return;
 906 }
 907
 908 void
 909 sfw_test_rpc_done (srpc_client_rpc_t *rpc)
 910 {
 911         sfw_test_unit_t     *tsu = rpc->crpc_priv;
 912         sfw_test_instance_t *tsi = tsu->tsu_instance;
 913         int                  done = 0;
 914
 915         tsi->tsi_ops->tso_done_rpc(tsu, rpc);
 916
 917         spin_lock(&tsi->tsi_lock);
 918
 919         LASSERT(sfw_test_active(tsi));
 920         LASSERT(!list_empty(&rpc->crpc_list));
 921
 922         list_del_init(&rpc->crpc_list);
 923
 924         /* batch is stopping or loop is done or get error */
 925         if (tsi->tsi_stopping ||
 926             tsu->tsu_loop == 0 ||
 927             (rpc->crpc_status != 0 && tsi->tsi_stoptsu_onerr))
 928                 done = 1;
 929
 930         /* dec ref for poster */
 931         srpc_client_rpc_decref(rpc);
 932
 933         spin_unlock(&tsi->tsi_lock);
 934
 935         if (!done) {
 936                 swi_schedule_workitem(&tsu->tsu_worker);
 937                 return;
 938         }
 939
 940         sfw_test_unit_done(tsu);
 941         return;
 942 }
 943
 944 int
 945 sfw_create_test_rpc(sfw_test_unit_t *tsu, lnet_process_id_t peer,
 946                     unsigned features, int nblk, int blklen,
 947                     srpc_client_rpc_t **rpcpp)
 948 {
 949         srpc_client_rpc_t   *rpc = NULL;
 950         sfw_test_instance_t *tsi = tsu->tsu_instance;
 951
 952         spin_lock(&tsi->tsi_lock);
 953
 954         LASSERT (sfw_test_active(tsi));
 955
 956         if (!list_empty(&tsi->tsi_free_rpcs)) {
 957                 /* pick request from buffer */
 958                 rpc = list_entry(tsi->tsi_free_rpcs.next,
 959                                  srpc_client_rpc_t, crpc_list);
 960                 LASSERT(nblk == rpc->crpc_bulk.bk_niov);
 961                 list_del_init(&rpc->crpc_list);
 962         }
 963
 964         spin_unlock(&tsi->tsi_lock);
 965
 966         if (rpc == NULL) {
 967                 rpc = srpc_create_client_rpc(peer, tsi->tsi_service, nblk,
 968                                              blklen, sfw_test_rpc_done,
 969                                              sfw_test_rpc_fini, tsu);
 970         } else {
 971                 srpc_init_client_rpc(rpc, peer, tsi->tsi_service, nblk,
 972                                      blklen, sfw_test_rpc_done,
 973                                      sfw_test_rpc_fini, tsu);
 974         }
 975
 976         if (rpc == NULL) {
 977                 CERROR("Can't create rpc for test %d\n", tsi->tsi_service);
 978                 return -ENOMEM;
 979         }
 980
 981         rpc->crpc_reqstmsg.msg_ses_feats = features;
 982         *rpcpp = rpc;
 983
 984         return 0;
 985 }
 986
 987 int
 988 sfw_run_test (swi_workitem_t *wi)
 989 {
 990         sfw_test_unit_t     *tsu = wi->swi_workitem.wi_data;
 991         sfw_test_instance_t *tsi = tsu->tsu_instance;
 992         srpc_client_rpc_t   *rpc = NULL;
 993
 994         LASSERT (wi == &tsu->tsu_worker);
 995
 996         if (tsi->tsi_ops->tso_prep_rpc(tsu, tsu->tsu_dest, &rpc) != 0) {
 997                 LASSERT (rpc == NULL);
 998                 goto test_done;
 999         }
1000
1001         LASSERT (rpc != NULL);
1002
1003         spin_lock(&tsi->tsi_lock);
1004
1005         if (tsi->tsi_stopping) {
1006                 list_add(&rpc->crpc_list, &tsi->tsi_free_rpcs);
1007                 spin_unlock(&tsi->tsi_lock);
1008                 goto test_done;
1009         }
1010
1011         if (tsu->tsu_loop > 0)
1012                 tsu->tsu_loop--;
1013
1014         list_add_tail(&rpc->crpc_list, &tsi->tsi_active_rpcs);
1015         spin_unlock(&tsi->tsi_lock);
1016
1017         spin_lock(&rpc->crpc_lock);
1018         rpc->crpc_timeout = rpc_timeout;
1019         srpc_post_rpc(rpc);
1020         spin_unlock(&rpc->crpc_lock);
1021         return 0;
1022
1023 test_done:
1024         /*
1025          * No one can schedule me now since:
1026          * - previous RPC, if any, has done and
1027          * - no new RPC is initiated.
1028          * - my batch is still active; no one can run it again now.
1029          * Cancel pending schedules and prevent future schedule attempts:
1030          */
1031         swi_exit_workitem(wi);
1032         sfw_test_unit_done(tsu);
1033         return 1;
1034 }
1035
1036 int
1037 sfw_run_batch (sfw_batch_t *tsb)
1038 {
1039         swi_workitem_t      *wi;
1040         sfw_test_unit_t     *tsu;
1041         sfw_test_instance_t *tsi;
1042
1043         if (sfw_batch_active(tsb)) {
1044                 CDEBUG(D_NET, "Batch already active: "LPU64" (%d)\n",
1045                        tsb->bat_id.bat_id, atomic_read(&tsb->bat_nactive));
1046                 return 0;
1047         }
1048
1049         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1050                 if (!tsi->tsi_is_client)        /* skip server instances */
1051                         continue;
1052
1053                 LASSERT(!tsi->tsi_stopping);
1054                 LASSERT(!sfw_test_active(tsi));
1055
1056                 atomic_inc(&tsb->bat_nactive);
1057
1058                 list_for_each_entry(tsu, &tsi->tsi_units, tsu_list) {
1059                         atomic_inc(&tsi->tsi_nactive);
1060                         tsu->tsu_loop = tsi->tsi_loop;
1061                         wi = &tsu->tsu_worker;
1062                         swi_init_workitem(wi, tsu, sfw_run_test,
1063                                           lst_sched_test[\
1064                                           lnet_cpt_of_nid(tsu->tsu_dest.nid)]);
1065                         swi_schedule_workitem(wi);
1066                 }
1067         }
1068
1069         return 0;
1070 }
1071
1072 int
1073 sfw_stop_batch (sfw_batch_t *tsb, int force)
1074 {
1075         sfw_test_instance_t *tsi;
1076         srpc_client_rpc_t   *rpc;
1077
1078         if (!sfw_batch_active(tsb)) {
1079                 CDEBUG(D_NET, "Batch "LPU64" inactive\n", tsb->bat_id.bat_id);
1080                 return 0;
1081         }
1082
1083         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1084                 spin_lock(&tsi->tsi_lock);
1085
1086                 if (!tsi->tsi_is_client ||
1087                     !sfw_test_active(tsi) || tsi->tsi_stopping) {
1088                         spin_unlock(&tsi->tsi_lock);
1089                         continue;
1090                 }
1091
1092                 tsi->tsi_stopping = 1;
1093
1094                 if (!force) {
1095                         spin_unlock(&tsi->tsi_lock);
1096                         continue;
1097                 }
1098
1099                 /* abort launched rpcs in the test */
1100                 list_for_each_entry(rpc, &tsi->tsi_active_rpcs, crpc_list) {
1101                         spin_lock(&rpc->crpc_lock);
1102
1103                         srpc_abort_rpc(rpc, -EINTR);
1104
1105                         spin_unlock(&rpc->crpc_lock);
1106                 }
1107
1108                 spin_unlock(&tsi->tsi_lock);
1109         }
1110
1111         return 0;
1112 }
1113
1114 int
1115 sfw_query_batch (sfw_batch_t *tsb, int testidx, srpc_batch_reply_t *reply)
1116 {
1117         sfw_test_instance_t *tsi;
1118
1119         if (testidx < 0)
1120                 return -EINVAL;
1121
1122         if (testidx == 0) {
1123                 reply->bar_active = atomic_read(&tsb->bat_nactive);
1124                 return 0;
1125         }
1126
1127         list_for_each_entry(tsi, &tsb->bat_tests, tsi_list) {
1128                 if (testidx-- > 1)
1129                         continue;
1130
1131                 reply->bar_active = atomic_read(&tsi->tsi_nactive);
1132                 return 0;
1133         }
1134
1135         return -ENOENT;
1136 }
1137
1138 void
1139 sfw_free_pages (srpc_server_rpc_t *rpc)
1140 {
1141         srpc_free_bulk(rpc->srpc_bulk);
1142         rpc->srpc_bulk = NULL;
1143 }
1144
1145 int
1146 sfw_alloc_pages(struct srpc_server_rpc *rpc, int cpt, int npages, int len,
1147                 int sink)
1148 {
1149         LASSERT(rpc->srpc_bulk == NULL);
1150         LASSERT(npages > 0 && npages <= LNET_MAX_IOV);
1151
1152         rpc->srpc_bulk = srpc_alloc_bulk(cpt, npages, len, sink);
1153         if (rpc->srpc_bulk == NULL)
1154                 return -ENOMEM;
1155
1156         return 0;
1157 }
1158
1159 int
1160 sfw_add_test (srpc_server_rpc_t *rpc)
1161 {
1162         sfw_session_t     *sn = sfw_data.fw_session;
1163         srpc_test_reply_t *reply = &rpc->srpc_replymsg.msg_body.tes_reply;
1164         srpc_test_reqst_t *request;
1165         int                rc;
1166         sfw_batch_t       *bat;
1167
1168         request = &rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst;
1169         reply->tsr_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1170
1171         if (request->tsr_loop == 0 ||
1172             request->tsr_concur == 0 ||
1173             request->tsr_sid.ses_nid == LNET_NID_ANY ||
1174             request->tsr_ndest > SFW_MAX_NDESTS ||
1175             (request->tsr_is_client && request->tsr_ndest == 0) ||
1176             request->tsr_concur > SFW_MAX_CONCUR ||
1177             request->tsr_service > SRPC_SERVICE_MAX_ID ||
1178             request->tsr_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID) {
1179                 reply->tsr_status = EINVAL;
1180                 return 0;
1181         }
1182
1183         if (sn == NULL || !sfw_sid_equal(request->tsr_sid, sn->sn_id) ||
1184             sfw_find_test_case(request->tsr_service) == NULL) {
1185                 reply->tsr_status = ENOENT;
1186                 return 0;
1187         }
1188
1189         bat = sfw_bid2batch(request->tsr_bid);
1190         if (bat == NULL) {
1191                 CERROR("dropping RPC %s from %s under memory pressure\n",
1192                         rpc->srpc_scd->scd_svc->sv_name,
1193                         libcfs_id2str(rpc->srpc_peer));
1194                 return -ENOMEM;
1195         }
1196
1197         if (sfw_batch_active(bat)) {
1198                 reply->tsr_status = EBUSY;
1199                 return 0;
1200         }
1201
1202         if (request->tsr_is_client && rpc->srpc_bulk == NULL) {
1203                 /* rpc will be resumed later in sfw_bulk_ready */
1204                 int     npg = sfw_id_pages(request->tsr_ndest);
1205                 int     len;
1206
1207                 if ((sn->sn_features & LST_FEAT_BULK_LEN) == 0) {
1208                         len = npg * PAGE_CACHE_SIZE;
1209
1210                 } else  {
1211                         len = sizeof(lnet_process_id_packed_t) *
1212                               request->tsr_ndest;
1213                 }
1214
1215                 return sfw_alloc_pages(rpc, CFS_CPT_ANY, npg, len, 1);
1216         }
1217
1218         rc = sfw_add_test_instance(bat, rpc);
1219         CDEBUG (rc == 0 ? D_NET : D_WARNING,
1220                 "%s test: sv %d %s, loop %d, concur %d, ndest %d\n",
1221                 rc == 0 ? "Added" : "Failed to add", request->tsr_service,
1222                 request->tsr_is_client ? "client" : "server",
1223                 request->tsr_loop, request->tsr_concur, request->tsr_ndest);
1224
1225         reply->tsr_status = (rc < 0) ? -rc : rc;
1226         return 0;
1227 }
1228
1229 int
1230 sfw_control_batch (srpc_batch_reqst_t *request, srpc_batch_reply_t *reply)
1231 {
1232         sfw_session_t *sn = sfw_data.fw_session;
1233         int            rc = 0;
1234         sfw_batch_t   *bat;
1235
1236         reply->bar_sid = (sn == NULL) ? LST_INVALID_SID : sn->sn_id;
1237
1238         if (sn == NULL || !sfw_sid_equal(request->bar_sid, sn->sn_id)) {
1239                 reply->bar_status = ESRCH;
1240                 return 0;
1241         }
1242
1243         bat = sfw_find_batch(request->bar_bid);
1244         if (bat == NULL) {
1245                 reply->bar_status = ENOENT;
1246                 return 0;
1247         }
1248
1249         switch (request->bar_opc) {
1250         case SRPC_BATCH_OPC_RUN:
1251                 rc = sfw_run_batch(bat);
1252                 break;
1253
1254         case SRPC_BATCH_OPC_STOP:
1255                 rc = sfw_stop_batch(bat, request->bar_arg);
1256                 break;
1257
1258         case SRPC_BATCH_OPC_QUERY:
1259                 rc = sfw_query_batch(bat, request->bar_testidx, reply);
1260                 break;
1261
1262         default:
1263                 return -EINVAL; /* drop it */
1264         }
1265
1266         reply->bar_status = (rc < 0) ? -rc : rc;
1267         return 0;
1268 }
1269
1270 int
1271 sfw_handle_server_rpc(struct srpc_server_rpc *rpc)
1272 {
1273         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1274         srpc_msg_t     *reply   = &rpc->srpc_replymsg;
1275         srpc_msg_t     *request = &rpc->srpc_reqstbuf->buf_msg;
1276         unsigned        features = LST_FEATS_MASK;
1277         int             rc = 0;
1278
1279         LASSERT(sfw_data.fw_active_srpc == NULL);
1280         LASSERT(sv->sv_id <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1281
1282         spin_lock(&sfw_data.fw_lock);
1283
1284         if (sfw_data.fw_shuttingdown) {
1285                 spin_unlock(&sfw_data.fw_lock);
1286                 return -ESHUTDOWN;
1287         }
1288
1289         /* Remove timer to avoid racing with it or expiring active session */
1290         if (sfw_del_session_timer() != 0) {
1291                 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1292                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1293                 spin_unlock(&sfw_data.fw_lock);
1294                 return -EAGAIN;
1295         }
1296
1297         sfw_data.fw_active_srpc = rpc;
1298         spin_unlock(&sfw_data.fw_lock);
1299
1300         sfw_unpack_message(request);
1301         LASSERT(request->msg_type == srpc_service2request(sv->sv_id));
1302
1303         /* rpc module should have checked this */
1304         LASSERT(request->msg_version == SRPC_MSG_VERSION);
1305
1306         if (sv->sv_id != SRPC_SERVICE_MAKE_SESSION &&
1307             sv->sv_id != SRPC_SERVICE_DEBUG) {
1308                 sfw_session_t *sn = sfw_data.fw_session;
1309
1310                 if (sn != NULL &&
1311                     sn->sn_features != request->msg_ses_feats) {
1312                         CNETERR("Features of framework RPC don't match "
1313                                 "features of current session: %x/%x\n",
1314                                 request->msg_ses_feats, sn->sn_features);
1315                         reply->msg_body.reply.status = EPROTO;
1316                         reply->msg_body.reply.sid    = sn->sn_id;
1317                         goto out;
1318                 }
1319
1320         } else if ((request->msg_ses_feats & ~LST_FEATS_MASK) != 0) {
1321                 /* NB: at this point, old version will ignore features and
1322                  * create new session anyway, so console should be able
1323                  * to handle this */
1324                 reply->msg_body.reply.status = EPROTO;
1325                 goto out;
1326         }
1327
1328         switch(sv->sv_id) {
1329         default:
1330                 LBUG ();
1331         case SRPC_SERVICE_TEST:
1332                 rc = sfw_add_test(rpc);
1333                 break;
1334
1335         case SRPC_SERVICE_BATCH:
1336                 rc = sfw_control_batch(&request->msg_body.bat_reqst,
1337                                        &reply->msg_body.bat_reply);
1338                 break;
1339
1340         case SRPC_SERVICE_QUERY_STAT:
1341                 rc = sfw_get_stats(&request->msg_body.stat_reqst,
1342                                    &reply->msg_body.stat_reply);
1343                 break;
1344
1345         case SRPC_SERVICE_DEBUG:
1346                 rc = sfw_debug_session(&request->msg_body.dbg_reqst,
1347                                        &reply->msg_body.dbg_reply);
1348                 break;
1349
1350         case SRPC_SERVICE_MAKE_SESSION:
1351                 rc = sfw_make_session(&request->msg_body.mksn_reqst,
1352                                       &reply->msg_body.mksn_reply);
1353                 break;
1354
1355         case SRPC_SERVICE_REMOVE_SESSION:
1356                 rc = sfw_remove_session(&request->msg_body.rmsn_reqst,
1357                                         &reply->msg_body.rmsn_reply);
1358                 break;
1359         }
1360
1361         if (sfw_data.fw_session != NULL)
1362                 features = sfw_data.fw_session->sn_features;
1363  out:
1364         reply->msg_ses_feats = features;
1365         rpc->srpc_done = sfw_server_rpc_done;
1366         spin_lock(&sfw_data.fw_lock);
1367
1368 #ifdef __KERNEL__
1369         if (!sfw_data.fw_shuttingdown)
1370                 sfw_add_session_timer();
1371 #else
1372         LASSERT(!sfw_data.fw_shuttingdown);
1373         sfw_add_session_timer();
1374 #endif
1375
1376         sfw_data.fw_active_srpc = NULL;
1377         spin_unlock(&sfw_data.fw_lock);
1378         return rc;
1379 }
1380
1381 int
1382 sfw_bulk_ready(struct srpc_server_rpc *rpc, int status)
1383 {
1384         struct srpc_service     *sv = rpc->srpc_scd->scd_svc;
1385         int                     rc;
1386
1387         LASSERT(rpc->srpc_bulk != NULL);
1388         LASSERT(sv->sv_id == SRPC_SERVICE_TEST);
1389         LASSERT(sfw_data.fw_active_srpc == NULL);
1390         LASSERT(rpc->srpc_reqstbuf->buf_msg.msg_body.tes_reqst.tsr_is_client);
1391
1392         spin_lock(&sfw_data.fw_lock);
1393
1394         if (status != 0) {
1395                 CERROR("Bulk transfer failed for RPC: "
1396                        "service %s, peer %s, status %d\n",
1397                        sv->sv_name, libcfs_id2str(rpc->srpc_peer), status);
1398                 spin_unlock(&sfw_data.fw_lock);
1399                 return -EIO;
1400         }
1401
1402         if (sfw_data.fw_shuttingdown) {
1403                 spin_unlock(&sfw_data.fw_lock);
1404                 return -ESHUTDOWN;
1405         }
1406
1407         if (sfw_del_session_timer() != 0) {
1408                 CERROR("dropping RPC %s from %s: racing with expiry timer\n",
1409                        sv->sv_name, libcfs_id2str(rpc->srpc_peer));
1410                 spin_unlock(&sfw_data.fw_lock);
1411                 return -EAGAIN;
1412         }
1413
1414         sfw_data.fw_active_srpc = rpc;
1415         spin_unlock(&sfw_data.fw_lock);
1416
1417         rc = sfw_add_test(rpc);
1418
1419         spin_lock(&sfw_data.fw_lock);
1420
1421 #ifdef __KERNEL__
1422         if (!sfw_data.fw_shuttingdown)
1423                 sfw_add_session_timer();
1424 #else
1425         LASSERT(!sfw_data.fw_shuttingdown);
1426         sfw_add_session_timer();
1427 #endif
1428
1429         sfw_data.fw_active_srpc = NULL;
1430         spin_unlock(&sfw_data.fw_lock);
1431         return rc;
1432 }
1433
1434 srpc_client_rpc_t *
1435 sfw_create_rpc(lnet_process_id_t peer, int service,
1436                unsigned features, int nbulkiov, int bulklen,
1437                void (*done)(srpc_client_rpc_t *), void *priv)
1438 {
1439         srpc_client_rpc_t *rpc = NULL;
1440
1441         spin_lock(&sfw_data.fw_lock);
1442
1443         LASSERT (!sfw_data.fw_shuttingdown);
1444         LASSERT (service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1445
1446         if (nbulkiov == 0 && !list_empty(&sfw_data.fw_zombie_rpcs)) {
1447                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1448                                      srpc_client_rpc_t, crpc_list);
1449                 list_del(&rpc->crpc_list);
1450
1451                 srpc_init_client_rpc(rpc, peer, service, 0, 0,
1452                                      done, sfw_client_rpc_fini, priv);
1453         }
1454
1455         spin_unlock(&sfw_data.fw_lock);
1456
1457         if (rpc == NULL) {
1458                 rpc = srpc_create_client_rpc(peer, service,
1459                                              nbulkiov, bulklen, done,
1460                                              nbulkiov != 0 ?  NULL :
1461                                              sfw_client_rpc_fini,
1462                                              priv);
1463         }
1464
1465         if (rpc != NULL) /* "session" is concept in framework */
1466                 rpc->crpc_reqstmsg.msg_ses_feats = features;
1467
1468         return rpc;
1469 }
1470
1471 void
1472 sfw_unpack_message (srpc_msg_t *msg)
1473 {
1474         if (msg->msg_magic == SRPC_MSG_MAGIC)
1475                 return; /* no flipping needed */
1476
1477         /* srpc module should guarantee I wouldn't get crap */
1478         LASSERT (msg->msg_magic == __swab32(SRPC_MSG_MAGIC));
1479
1480         if (msg->msg_type == SRPC_MSG_STAT_REQST) {
1481                 srpc_stat_reqst_t *req = &msg->msg_body.stat_reqst;
1482
1483                 __swab32s(&req->str_type);
1484                 __swab64s(&req->str_rpyid);
1485                 sfw_unpack_sid(req->str_sid);
1486                 return;
1487         }
1488
1489         if (msg->msg_type == SRPC_MSG_STAT_REPLY) {
1490                 srpc_stat_reply_t *rep = &msg->msg_body.stat_reply;
1491
1492                 __swab32s(&rep->str_status);
1493                 sfw_unpack_sid(rep->str_sid);
1494                 sfw_unpack_fw_counters(rep->str_fw);
1495                 sfw_unpack_rpc_counters(rep->str_rpc);
1496                 sfw_unpack_lnet_counters(rep->str_lnet);
1497                 return;
1498         }
1499
1500         if (msg->msg_type == SRPC_MSG_MKSN_REQST) {
1501                 srpc_mksn_reqst_t *req = &msg->msg_body.mksn_reqst;
1502
1503                 __swab64s(&req->mksn_rpyid);
1504                 __swab32s(&req->mksn_force);
1505                 sfw_unpack_sid(req->mksn_sid);
1506                 return;
1507         }
1508
1509         if (msg->msg_type == SRPC_MSG_MKSN_REPLY) {
1510                 srpc_mksn_reply_t *rep = &msg->msg_body.mksn_reply;
1511
1512                 __swab32s(&rep->mksn_status);
1513                 __swab32s(&rep->mksn_timeout);
1514                 sfw_unpack_sid(rep->mksn_sid);
1515                 return;
1516         }
1517
1518         if (msg->msg_type == SRPC_MSG_RMSN_REQST) {
1519                 srpc_rmsn_reqst_t *req = &msg->msg_body.rmsn_reqst;
1520
1521                 __swab64s(&req->rmsn_rpyid);
1522                 sfw_unpack_sid(req->rmsn_sid);
1523                 return;
1524         }
1525
1526         if (msg->msg_type == SRPC_MSG_RMSN_REPLY) {
1527                 srpc_rmsn_reply_t *rep = &msg->msg_body.rmsn_reply;
1528
1529                 __swab32s(&rep->rmsn_status);
1530                 sfw_unpack_sid(rep->rmsn_sid);
1531                 return;
1532         }
1533
1534         if (msg->msg_type == SRPC_MSG_DEBUG_REQST) {
1535                 srpc_debug_reqst_t *req = &msg->msg_body.dbg_reqst;
1536
1537                 __swab64s(&req->dbg_rpyid);
1538                 __swab32s(&req->dbg_flags);
1539                 sfw_unpack_sid(req->dbg_sid);
1540                 return;
1541         }
1542
1543         if (msg->msg_type == SRPC_MSG_DEBUG_REPLY) {
1544                 srpc_debug_reply_t *rep = &msg->msg_body.dbg_reply;
1545
1546                 __swab32s(&rep->dbg_nbatch);
1547                 __swab32s(&rep->dbg_timeout);
1548                 sfw_unpack_sid(rep->dbg_sid);
1549                 return;
1550         }
1551
1552         if (msg->msg_type == SRPC_MSG_BATCH_REQST) {
1553                 srpc_batch_reqst_t *req = &msg->msg_body.bat_reqst;
1554
1555                 __swab32s(&req->bar_opc);
1556                 __swab64s(&req->bar_rpyid);
1557                 __swab32s(&req->bar_testidx);
1558                 __swab32s(&req->bar_arg);
1559                 sfw_unpack_sid(req->bar_sid);
1560                 __swab64s(&req->bar_bid.bat_id);
1561                 return;
1562         }
1563
1564         if (msg->msg_type == SRPC_MSG_BATCH_REPLY) {
1565                 srpc_batch_reply_t *rep = &msg->msg_body.bat_reply;
1566
1567                 __swab32s(&rep->bar_status);
1568                 sfw_unpack_sid(rep->bar_sid);
1569                 return;
1570         }
1571
1572         if (msg->msg_type == SRPC_MSG_TEST_REQST) {
1573                 srpc_test_reqst_t *req = &msg->msg_body.tes_reqst;
1574
1575                 __swab64s(&req->tsr_rpyid);
1576                 __swab64s(&req->tsr_bulkid);
1577                 __swab32s(&req->tsr_loop);
1578                 __swab32s(&req->tsr_ndest);
1579                 __swab32s(&req->tsr_concur);
1580                 __swab32s(&req->tsr_service);
1581                 sfw_unpack_sid(req->tsr_sid);
1582                 __swab64s(&req->tsr_bid.bat_id);
1583                 return;
1584         }
1585
1586         if (msg->msg_type == SRPC_MSG_TEST_REPLY) {
1587                 srpc_test_reply_t *rep = &msg->msg_body.tes_reply;
1588
1589                 __swab32s(&rep->tsr_status);
1590                 sfw_unpack_sid(rep->tsr_sid);
1591                 return;
1592         }
1593
1594         if (msg->msg_type == SRPC_MSG_JOIN_REQST) {
1595                 srpc_join_reqst_t *req = &msg->msg_body.join_reqst;
1596
1597                 __swab64s(&req->join_rpyid);
1598                 sfw_unpack_sid(req->join_sid);
1599                 return;
1600         }
1601
1602         if (msg->msg_type == SRPC_MSG_JOIN_REPLY) {
1603                 srpc_join_reply_t *rep = &msg->msg_body.join_reply;
1604
1605                 __swab32s(&rep->join_status);
1606                 __swab32s(&rep->join_timeout);
1607                 sfw_unpack_sid(rep->join_sid);
1608                 return;
1609         }
1610
1611         LBUG ();
1612         return;
1613 }
1614
1615 void
1616 sfw_abort_rpc (srpc_client_rpc_t *rpc)
1617 {
1618         LASSERT(atomic_read(&rpc->crpc_refcount) > 0);
1619         LASSERT(rpc->crpc_service <= SRPC_FRAMEWORK_SERVICE_MAX_ID);
1620
1621         spin_lock(&rpc->crpc_lock);
1622         srpc_abort_rpc(rpc, -EINTR);
1623         spin_unlock(&rpc->crpc_lock);
1624         return;
1625 }
1626
1627 void
1628 sfw_post_rpc (srpc_client_rpc_t *rpc)
1629 {
1630         spin_lock(&rpc->crpc_lock);
1631
1632         LASSERT(!rpc->crpc_closed);
1633         LASSERT(!rpc->crpc_aborted);
1634         LASSERT(list_empty(&rpc->crpc_list));
1635         LASSERT(!sfw_data.fw_shuttingdown);
1636
1637         rpc->crpc_timeout = rpc_timeout;
1638         srpc_post_rpc(rpc);
1639
1640         spin_unlock(&rpc->crpc_lock);
1641         return;
1642 }
1643
1644 static srpc_service_t sfw_services[] =
1645 {
1646         {
1647                 /* sv_id */    SRPC_SERVICE_DEBUG,
1648                 /* sv_name */  "debug",
1649                 0
1650         },
1651         {
1652                 /* sv_id */    SRPC_SERVICE_QUERY_STAT,
1653                 /* sv_name */  "query stats",
1654                 0
1655         },
1656         {
1657                 /* sv_id */    SRPC_SERVICE_MAKE_SESSION,
1658                 /* sv_name */  "make session",
1659                 0
1660         },
1661         {
1662                 /* sv_id */    SRPC_SERVICE_REMOVE_SESSION,
1663                 /* sv_name */  "remove session",
1664                 0
1665         },
1666         {
1667                 /* sv_id */    SRPC_SERVICE_BATCH,
1668                 /* sv_name */  "batch service",
1669                 0
1670         },
1671         {
1672                 /* sv_id */    SRPC_SERVICE_TEST,
1673                 /* sv_name */  "test service",
1674                 0
1675         },
1676         {
1677                 /* sv_id */    0,
1678                 /* sv_name */  NULL,
1679                 0
1680         }
1681 };
1682
1683 extern sfw_test_client_ops_t ping_test_client;
1684 extern srpc_service_t        ping_test_service;
1685 extern void ping_init_test_client(void);
1686 extern void ping_init_test_service(void);
1687
1688 extern sfw_test_client_ops_t brw_test_client;
1689 extern srpc_service_t        brw_test_service;
1690 extern void brw_init_test_client(void);
1691 extern void brw_init_test_service(void);
1692
1693
1694 int
1695 sfw_startup (void)
1696 {
1697         int              i;
1698         int              rc;
1699         int              error;
1700         srpc_service_t  *sv;
1701         sfw_test_case_t *tsc;
1702
1703 #ifndef __KERNEL__
1704         char *s;
1705
1706         s = getenv("SESSION_TIMEOUT");
1707         session_timeout = s != NULL ? atoi(s) : session_timeout;
1708
1709         s = getenv("RPC_TIMEOUT");
1710         rpc_timeout = s != NULL ? atoi(s) : rpc_timeout;
1711 #endif
1712
1713         if (session_timeout < 0) {
1714                 CERROR ("Session timeout must be non-negative: %d\n",
1715                         session_timeout);
1716                 return -EINVAL;
1717         }
1718
1719         if (rpc_timeout < 0) {
1720                 CERROR ("RPC timeout must be non-negative: %d\n",
1721                         rpc_timeout);
1722                 return -EINVAL;
1723         }
1724
1725         if (session_timeout == 0)
1726                 CWARN ("Zero session_timeout specified "
1727                        "- test sessions never expire.\n");
1728
1729         if (rpc_timeout == 0)
1730                 CWARN ("Zero rpc_timeout specified "
1731                        "- test RPC never expire.\n");
1732
1733         memset(&sfw_data, 0, sizeof(struct smoketest_framework));
1734
1735         sfw_data.fw_session     = NULL;
1736         sfw_data.fw_active_srpc = NULL;
1737         spin_lock_init(&sfw_data.fw_lock);
1738         atomic_set(&sfw_data.fw_nzombies, 0);
1739         INIT_LIST_HEAD(&sfw_data.fw_tests);
1740         INIT_LIST_HEAD(&sfw_data.fw_zombie_rpcs);
1741         INIT_LIST_HEAD(&sfw_data.fw_zombie_sessions);
1742
1743         brw_init_test_client();
1744         brw_init_test_service();
1745         rc = sfw_register_test(&brw_test_service, &brw_test_client);
1746         LASSERT (rc == 0);
1747
1748         ping_init_test_client();
1749         ping_init_test_service();
1750         rc = sfw_register_test(&ping_test_service, &ping_test_client);
1751         LASSERT (rc == 0);
1752
1753         error = 0;
1754         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1755                 sv = tsc->tsc_srv_service;
1756
1757                 rc = srpc_add_service(sv);
1758                 LASSERT(rc != -EBUSY);
1759                 if (rc != 0) {
1760                         CWARN("Failed to add %s service: %d\n",
1761                               sv->sv_name, rc);
1762                         error = rc;
1763                 }
1764         }
1765
1766         for (i = 0; ; i++) {
1767                 sv = &sfw_services[i];
1768                 if (sv->sv_name == NULL) break;
1769
1770                 sv->sv_bulk_ready = NULL;
1771                 sv->sv_handler    = sfw_handle_server_rpc;
1772                 sv->sv_wi_total   = SFW_FRWK_WI_MAX;
1773                 if (sv->sv_id == SRPC_SERVICE_TEST)
1774                         sv->sv_bulk_ready = sfw_bulk_ready;
1775
1776                 rc = srpc_add_service(sv);
1777                 LASSERT (rc != -EBUSY);
1778                 if (rc != 0) {
1779                         CWARN ("Failed to add %s service: %d\n",
1780                                sv->sv_name, rc);
1781                         error = rc;
1782                 }
1783
1784                 /* about to sfw_shutdown, no need to add buffer */
1785                 if (error) continue;
1786
1787                 rc = srpc_service_add_buffers(sv, sv->sv_wi_total);
1788                 if (rc != 0) {
1789                         CWARN("Failed to reserve enough buffers: "
1790                               "service %s, %d needed: %d\n",
1791                               sv->sv_name, sv->sv_wi_total, rc);
1792                         error = -ENOMEM;
1793                 }
1794         }
1795
1796         if (error != 0)
1797                 sfw_shutdown();
1798         return error;
1799 }
1800
1801 void
1802 sfw_shutdown (void)
1803 {
1804         srpc_service_t  *sv;
1805         sfw_test_case_t *tsc;
1806         int              i;
1807
1808         spin_lock(&sfw_data.fw_lock);
1809
1810         sfw_data.fw_shuttingdown = 1;
1811 #ifdef __KERNEL__
1812         lst_wait_until(sfw_data.fw_active_srpc == NULL, sfw_data.fw_lock,
1813                        "waiting for active RPC to finish.\n");
1814 #else
1815         LASSERT (sfw_data.fw_active_srpc == NULL);
1816 #endif
1817
1818         if (sfw_del_session_timer() != 0)
1819                 lst_wait_until(sfw_data.fw_session == NULL, sfw_data.fw_lock,
1820                                "waiting for session timer to explode.\n");
1821
1822         sfw_deactivate_session();
1823         lst_wait_until(atomic_read(&sfw_data.fw_nzombies) == 0,
1824                        sfw_data.fw_lock,
1825                        "waiting for %d zombie sessions to die.\n",
1826                        atomic_read(&sfw_data.fw_nzombies));
1827
1828         spin_unlock(&sfw_data.fw_lock);
1829
1830         for (i = 0; ; i++) {
1831                 sv = &sfw_services[i];
1832                 if (sv->sv_name == NULL)
1833                         break;
1834
1835                 srpc_shutdown_service(sv);
1836                 srpc_remove_service(sv);
1837         }
1838
1839         list_for_each_entry(tsc, &sfw_data.fw_tests, tsc_list) {
1840                 sv = tsc->tsc_srv_service;
1841                 srpc_shutdown_service(sv);
1842                 srpc_remove_service(sv);
1843         }
1844
1845         while (!list_empty(&sfw_data.fw_zombie_rpcs)) {
1846                 srpc_client_rpc_t *rpc;
1847
1848                 rpc = list_entry(sfw_data.fw_zombie_rpcs.next,
1849                                  srpc_client_rpc_t, crpc_list);
1850                 list_del(&rpc->crpc_list);
1851
1852                 LIBCFS_FREE(rpc, srpc_client_rpc_size(rpc));
1853         }
1854
1855         for (i = 0; ; i++) {
1856                 sv = &sfw_services[i];
1857                 if (sv->sv_name == NULL)
1858                         break;
1859
1860                 srpc_wait_service_shutdown(sv);
1861         }
1862
1863         while (!list_empty(&sfw_data.fw_tests)) {
1864                 tsc = list_entry(sfw_data.fw_tests.next,
1865                                  sfw_test_case_t, tsc_list);
1866
1867                 srpc_wait_service_shutdown(tsc->tsc_srv_service);
1868
1869                 list_del(&tsc->tsc_list);
1870                 LIBCFS_FREE(tsc, sizeof(*tsc));
1871         }
1872
1873         return;
1874 }